GlassBallAI: A Dataset for Studying LLM Market Prediction Drift with Gemini Pro 2.5
Summary
GlassBallAI provides a dataset of LLM-generated stock predictions produced with Gemini Pro 2.5 to study how model reasoning and confidence drift over time. It emphasizes reproducibility through fixed prompts, checkpoint structure, and ground-truth context, making it a valuable resource for AI tool evaluation and content creators exploring model behavior under uncertainty. The dataset is non-commercial and intended for research use, not trading.